The News, Explained

Kakao Mobility announced a November 9 memorandum of understanding with NVIDIA to develop Level 4 autonomous driving technology. The initial plan is to integrate its own driving AI with NVIDIA computing, sensing and safety technology in Kia vehicles, then pursue development and supply of vehicles for autonomous services. The announcement’s image caption identifies the Kia PV5 as the first vehicle intended to receive the technology. Source

Level 4 means the system performs the driving within a defined operating area, while occupants do not need to operate the vehicle. The US National Highway Traffic Safety Administration, or NHTSA, explicitly describes limited service areas. Understanding a Level 4 vehicle therefore also means understanding where and under what conditions it can operate. Source

Kakao Mobility plans to develop its E2E, or end-to-end, driving AI, which connects perception of the surroundings through to driving decisions in one model. To handle AI computation in real time, it plans to introduce an autonomous driving computing unit containing two NVIDIA DRIVE AGX Thor units. The computing unit is the equipment that processes the AI workload inside the vehicle. Source

Hyperion is a reference architecture for combining sensors and computing equipment. Kakao Mobility plans to use it to optimize compatibility and stable operation between the vehicle, sensors and computing unit. NVIDIA also describes Hyperion as a sensor and compute platform. This connects developing an AI model with integrating it into the equipment of a working vehicle. Source Source

The partners plan to apply the Halos safety system to driving software development and validation. NVIDIA describes it as spanning vehicle architecture, system software and driving algorithms. Kakao Mobility’s ambition for high safety remains a goal for the development and validation work ahead. The announcement does not specify a passenger service launch date, fares or operating area. Source Source

OYOPICK’s Take

The interesting approach here is retaining an in-house driving AI while connecting sensors, computing and safety validation at the vehicle level. Passengers need an explanation of how that combination works in actual operating conditions more than a list of component specifications. We hope the next demonstrations disclose concrete operating areas and conditions, responses to system faults, and the scope of validation. If that evidence accumulates and leads to a service, the project could also develop into a wider choice of travel for people who find driving difficult. Source Source